aiLL 7/7/23 - 2 - The Most Important AI Tool in Human History?

Description

Chapters

Transcript

0:16 yeah all right I'm coming back yep is
0:19 back welcome yep yep welcome
0:22 XXX all right people are coming back if
0:25 you would go ahead thank you very much
0:27 so CN is so was was that just some un
0:30 trolling me that said that cuz I've been
0:32 booted here twice but it just booted me
0:35 it just booted me and then said you
0:37 can't come back for 7 days I never got
0:38 anything like that okay all right that's
0:41 fine I figured that but like because I'm
0:43 so far behind on my things like I didn't
0:45 know what it was oh you know those are
0:47 trolls okay good great all right thank
0:49 you people you know it's again this is
0:52 this is like you know remedial remedial
0:55 old guy Tick Tock
0:59 lessons
1:03 anyway go ahead and tap the screen uh
1:05 let's get some likes going and maybe you
1:06 can share the live so so people can do
1:08 it thanks for the rose that's great um
1:12 you know those are trolls no I didn't
1:13 know those were trolls because it it
1:15 looked like it had a Tik Tock um icon
1:19 but I guess someone could just put a Tik
1:21 Tock icon there yeah I get
1:24 it whatever what are you going to do we
1:28 we'll we'll have a we'll have a more
1:29 intimate grouping um cuz cuz what did we
1:32 get up to we were up to like 130 people
1:34 there for a while so maybe we'll get
1:36 some more people go ahead and tap the
1:37 screen share the live let me know
1:40 encourage more viewers to share yeah
1:43 viewers share would you would you share
1:45 would you do that would you would you
1:48 help an old man all I've ever wanted was
1:51 a Tik Tock following I came on the Tik
1:55 Tok I I just wanted to be loved I just
1:59 wanted to be loved d Can I Be Loved
2:02 won't you share this live oh yeah
2:04 there's like 163 people here holy crap
2:07 okay when I'm talking like an idiot it's
2:09 one of the things that makes me happy
2:11 that people are like scrolling through
2:12 Tik Tok and there's just this weird ass
2:14 old man going ah yes I was hoping that
2:17 people would like me
2:22 someday but anyway thank you um oh I'm
2:25 supposed to call it the Ticky talk yeah
2:27 welcome to the Chickity talk Sally
2:31 Sally how do I turn off the tick
2:35 talk I see my face again what do I do I
2:39 I I don't want to see my face it scares
2:41 have you seen my hair Sally Sally yeah
2:45 exactly all right love your energy thank
2:47 you very
2:48 much you'll be permanently banned that's
2:51 a shitty troll all
2:54 right cuz I didn't do anything and if
2:56 they wanted to if they were going to
2:58 boot me they had just booted me all
2:59 right I'll get there I'll figure this
3:01 out anyway tell me is prompt engineering
3:03 an actual job it is and it's it is and
3:07 it's not and it's going to be and it's
3:09 not so um it's a job for any company
3:14 trying to figure out what the [ __ ] to do
3:16 with AI um they need people that
3:18 understand how to talk to these tools
3:20 and right now these tools are
3:24 so the troll one at least temporarily I
3:27 know let's go back to the content what
3:29 content was I doing
3:31 tell me what content I was doing I'll go
3:33 back to it cuz cuz you know I got thrown
3:35 I was like I was like you're going to
3:37 pay me burm permanently I can't lose my
3:39 Ticky talkie Ticky
3:45 talky Elder talk this is it
3:49 welcome old guys trying to
3:55 twerk that notification that'll get you
3:58 banned is a troll account those son of a
4:00 [ __ ] sons of
4:03 [ __ ] uh we can go back to prompt
4:05 engineering is a job let's go back to
4:06 the content I don't know what the
4:07 content was to tell me what the content
4:10 was was there content I know that's
4:13 where I am Amelia's wife I don't think
4:14 there was content you where you talk
4:17 about the cost oh oh okay so I was
4:20 talking
4:21 about okay so so if you're new here my
4:24 name is Kyle Shannon this is the AI
4:26 learning lab uh these URLs here are just
4:29 here really for for your it's like a
4:32 public service announcement if you
4:33 haven't played with chat GPT go to chat.
4:36 open.com that's the official website of
4:39 chat GPT just go there and start playing
4:41 and if you don't know what to do there
4:42 go to prompt. chat and that's a document
4:45 that will tell you how to do it I'm not
4:47 affiliated with them I don't make money
4:48 I don't give a [ __ ] anyway okay so what
4:51 we were talking about before was you
4:53 know someone asked about like you know
4:55 how much is the you know what do I think
4:57 the server cost is or the the like the
4:59 hosting cost
5:00 um it's tremendously expensive to to do
5:05 for for us to be able to play with these
5:07 like it is we are in a
5:10 very very very very lucky position as
5:15 non people working at these companies to
5:19 have access to these tools
5:22 um the fact that chat GPT went from Z to
5:27 100 million users in 6 weeks starting
5:30 last November changed everything
5:34 right Microsoft was not expecting it to
5:37 be that popular open AI was not
5:39 expecting it to be that popular Google
5:42 was certainly not expecting it to be
5:44 that popular but but when it did what it
5:47 did everyone started taking this
5:50 seriously Microsoft went from a billion
5:52 dollars to1 billion doar and and
5:54 basically basically said we're going to
5:57 support this
5:58 thing and we're going to the the
6:02 interesting thing about Microsoft as
6:04 compared to Google so Google all they
6:06 have is their ad Revenue they've got you
6:08 know they've got their search ad Revenue
6:11 Microsoft has their
6:13 entire software product Suite that
6:16 they're incorporating chat GPT into so
6:19 they' got a completely different way to
6:20 monetize it and the CEO of Microsoft has
6:22 said he's willing to lose money on chat
6:26 GPT as part of Bing to [ __ ] with Google
6:29 and he said it with some Glee he didn't
6:31 use those words but you know that's what
6:33 he was saying um and also they're going
6:35 to be able to monetize this through you
6:37 know an additional monthly charge for
6:39 Office 365 and for Windows and for all
6:43 the Azure cloud services so they're
6:45 they're going to make their money back
6:46 on the business services and the
6:47 Consulting and all that other [ __ ] that
6:49 they do so they can take a loss leader
6:51 on this front but the fact that it got
6:53 so popular is why everyone is is sort of
6:57 piling onto this AI stuff why it's got
7:00 so much hype um it's also I think the
7:03 reason that meta let their llama model
7:07 out into the wild um really interesting
7:10 move there they they said it was a
7:12 mistake you know it's a leak it was it
7:14 wasn't really a leak they're like well
7:15 we give it to a few developers we didn't
7:17 think they'd give it to everyone what
7:19 are we to do um so but I think that was
7:22 on purpose but all of this is is all of
7:24 these companies now going we've got AI
7:26 we've got AI oh we're going to be
7:28 leaders too we've got a big model we've
7:29 got a bigger model all that Panic is
7:32 because of chat gbt but the question was
7:36 like how expensive is it really [ __ ]
7:39 expensive now one of the cool things
7:42 about the fact that meta released all
7:45 their large language models that like
7:47 the the base called what what are called
7:49 the weights basically the The Secret
7:51 Sauce the fact that they release those
7:54 into the world the developer Community
7:56 The open- Source Community is pouncing
7:59 all over these things things and they're
8:01 now able to train for about $100 in
8:04 compute cost and get results almost as
8:07 good as Google doing a foundational
8:09 large language model that takes like 6
8:11 months and $10 million the developer
8:13 Community is getting those things to
8:16 similar levels of performance for like a
8:18 hundred bucks so it's like it's [ __ ]
8:20 wild it is insane right now it really is
8:24 just nuts so anyway um if if you're new
8:28 here uh put any questions or comments or
8:32 thoughts or maybe even something you're
8:33 trying to work on in the comments below
8:35 I'll do my best to answer them I'm going
8:37 to flip cameras here and pause so I
8:39 don't break but I'll show you these oh
8:42 we don't have that many people here so
8:44 um go play with those Ur URLs over there
8:48 um while you listen to me ramble if you
8:50 haven't played with chat GPT it's
8:52 actually really important for all of us
8:55 and I like I believe that deep deep in
8:57 my
8:58 soul uh when will office when will
9:02 Microsoft integrate GPT into office I I
9:05 would assume as soon as they're ready
9:06 like like Microsoft is
9:10 is they're [ __ ] acting like apple
9:13 like like like I've hated Microsoft for
9:16 decades like I was a Microsoft Basher
9:18 for [ __ ] decades we used to make fun
9:20 of them in the early days of the
9:22 worldwide web I had this group I started
9:24 called the worldwide web artist
9:25 Consortium and we would meet at the Sony
9:29 headquarters in New York City and and
9:31 Sony Sony gave us this it was this
9:34 conference it was this boardroom but it
9:36 sat like 200 people it was this massive
9:39 Square table and we'd all sit around
9:41 this table it was like a hundred people
9:43 would sit around the table and another
9:45 hundred were around them and we would
9:47 have industry people come in and they
9:49 would present like new web Technologies
9:52 and every time Microsoft came in we just
9:55 [ __ ] annihilated them [ __ ]
9:57 annihilated them they were just they
10:00 were just hated but but what they're
10:01 doing with with this chat GPT and with
10:04 with open AI is it's it's [ __ ]
10:06 brilliant um so probably as soon as they
10:09 can um being Microsoft they're not going
10:12 to they're not going to release
10:14 something that hasn't been really well
10:15 thought out and I think how you actually
10:18 integrate the co-pilot into Excel is
10:22 going to be different than how you do it
10:23 in word and and you know is it always
10:26 there is it like clippy that just
10:28 [ __ ] you know shows up annoys you
10:30 like like they they seem to I I don't
10:33 know what's going on with their product
10:36 development over there but their their
10:39 product development seem seems to have
10:41 gotten sort of a a step change better
10:45 than it's ever been like like their
10:47 interface and their product has always
10:49 been just over bloated horseshit right
10:54 we all have to use it cuz it all does
10:55 what we needed to do but like it's just
10:57 been bad what's what's beenen happening
10:59 lately is is they've been introducing
11:01 these new tools like it feels like
11:03 they've got some new product person
11:05 there or product team there that's
11:07 that's dramatically improving things so
11:09 as soon as it's ready I'm sure they they
11:11 they want to get it out before Google
11:14 does and I think they will so I'm
11:20 guessing fall you know September October
11:24 I don't know I don't have a [ __ ] clue
11:26 but I want it I wanted code interpreter
11:28 and we got code interpreter
11:32 yeah how much do it cost for all these
11:34 big for all these big computers like
11:37 like staggering amounts of money again
11:39 oh that oh that was the thing I was
11:41 saying so Microsoft you know they went
11:44 from 1 billion doll to1
11:47 billion and then Sam Alman has said to
11:50 get to AGI artificial general
11:52 intelligence was that that's the goal
11:53 for open AI is to get to a system that's
11:57 smarter than humans that can reason and
11:59 you know basically be smarter than than
12:01 the smartest
12:02 human that's their goal and Sam Alman
12:05 said he thinks it's going to take them a
12:06 hundred billion doar to get there and
12:08 and that's going to be with you know all
12:10 of these tools out in the world and
12:12 being used so it's it's going to be a
12:14 lot it's a lot of money a lot of money
12:16 how can I stay active on
12:18 teams I don't know show up to
12:22 practices kick the ball do laps do
12:26 push-ups Oh you mean Microsoft teams
12:29 I don't know log in I don't know what
12:31 you're talking
12:37 about I am looking forward to open AI in
12:41 the XX what's the
12:44 XX cross
12:49 pixs that
12:51 is wait that is me that dude I don't
12:54 know what you're saying there I don't
12:56 what is XX is that some something that
12:59 an old guy wouldn't understand are you
13:01 trolling an old guy by using youngster
13:05 terms you son of a [ __ ]
13:08 I all right we need more likes to boost
13:12 live guys support by tapping the screen
13:14 oh thank you very much yeah feel free to
13:16 tap feel free to share the live anything
13:18 we can do to let people know are here um
13:20 I really do think it's important so so I
13:23 know I'm a little smarmy about this and
13:24 I've been doing this for a while so I'm
13:26 getting a little horse and you know numb
13:29 buted
13:31 um and and a little Punchy um but but
13:35 the the the real point of this channel
13:37 is to demystify this AI stuff and get
13:40 people to try it get people excited
13:43 about it and not excited about it like
13:45 you have to love it like I don't give a
13:47 [ __ ] if you love it or hate it but I
13:49 don't want you to hate it if you've
13:50 never tried it like the whole point here
13:53 is learn this [ __ ] enough to at least
13:58 have a Kevin McAllister moment which for
14:01 young people there's this movie that us
14:03 old people saw when we were younger
14:05 called home alone and Kevin McAllister
14:08 did this
14:09 once and when you use these
14:13 tools at some point it's going to do
14:15 something that makes you do that and you
14:18 owe it to yourself to use these tools
14:21 until you've had that moment and then at
14:24 that point you can say [ __ ] it this is
14:27 all [ __ ] right I think I think the
14:29 robots are going to kill us I'm just
14:31 going to sit back I'm moving to the
14:33 beach um or you'll go the other way and
14:36 you'll go oh holy [ __ ] [ __ ] like uh
14:39 does this mean I should probably start
14:41 learning this like right now
14:44 yeah because AI is not going to take
14:46 your job but someone using AI
14:49 will every job every knowledge worker
14:52 job every creative job I I I can't
14:58 foresee a job that uses computers that
15:01 is not going to be profoundly impacted
15:03 by this [ __ ] and many jobs that don't
15:06 even use computers that are impacted by
15:08 people using computers profound profound
15:11 changes are coming from this so that's
15:12 what this channel is all
15:14 about um what's nuts I'm
15:17 nuts hey there hope you're having a
15:20 great evening thank you very much
15:21 silhouette I appreciate that do you
15:23 think oh Tik Tock just jumped me to the
15:26 bottom oh tick tock
15:30 oh here we go there's a danger of using
15:32 this do I think there is a danger of
15:34 using this in the future danger of using
15:37 what in the future AI oh like is there a
15:40 danger of using AI yeah it's
15:43 so 3 days ago open AI made an
15:46 announcement that they're dedicating 20%
15:50 of their staff's effort that's a lot for
15:53 for a company that's doing as much as
15:55 they are they're dedicating 20% of their
15:58 staff effort to what they're calling
16:00 super
16:01 alignment alignment in AI
16:06 means are the goals of the AI aligned
16:10 with human
16:14 goals cuz if they're not and this [ __ ]
16:18 gets smarter than
16:20 us and it and it doesn't have alignment
16:23 and it's like you
16:24 know I know those humans want me to not
16:28 shut down nuclear power plants and I
16:31 know they want me to not do this but I
16:34 think it might be in my best interest
16:36 for my survival as a system to do
16:39 whatever it's going to do so we don't
16:41 want that so so so openai if you want to
16:46 read about it you can go to um
16:48 open.com blog and I think it's it's it's
16:52 either the last thing they did or the
16:53 one before but it's this thing on super
16:55 alignment you can go read about it it's
16:56 it's fascinating but they're basically
16:58 going to use chat GPT to do all sorts of
17:00 simulations and and do all these things
17:02 to make sure that these tools are
17:04 aligned with our values as a human
17:07 species so there is a danger because
17:10 these things when they start executing
17:12 code which code interpreter executes
17:15 code now it's got safety guard rails and
17:17 things like that but again you've got
17:19 this whole open source community so
17:20 there's a whole Community out there with
17:22 these large language models without
17:24 safety guard rails on them right so
17:26 that's happening in parallel with a
17:28 company like open AI saying hey we're
17:30 going to dedicate 20% of our staff to
17:32 the safety piece piece of this so um yes
17:37 it's possible but I also think that
17:39 another reason for the channel why why I
17:42 do these and why I'm so passionate about
17:43 it
17:44 is I firmly believe
17:49 that even small individual efforts on
17:53 all of our parts to do something good
17:55 with this technology anything we do to
17:57 bend it toward the light Bend end it
17:59 toward the good right now is going to
18:01 have a disproportionately positive
18:03 impact on the direction of this stuff in
18:05 the future if we all just sit back and
18:07 go ah I've got three years I don't
18:10 really need it's it's probably hype
18:12 someone told me about blockchain and
18:14 nfts and I I lost my grandma's money and
18:17 so this is probably just one of those
18:19 anyone that's the more people that sit
18:21 back and just do that and dismiss it and
18:24 and kind of live in fear of it um it
18:27 ain't helping
18:29 right but everyone that goes they have
18:32 their Kevin McAllister moment they're
18:34 holy [ __ ] I can do that can it do my job
18:38 here it oh [ __ ] what do I do now you
18:42 learn you get curious you start
18:44 exploring oh well maybe I could do this
18:46 maybe I could do that and then people
18:48 have these breakthroughs where they're
18:49 like oh I know what I could do and I'm
18:52 I'm seeing it consistantly people are
18:53 getting their heads around this [ __ ] and
18:55 they're Reinventing themselves and
18:56 they're like making these really cool
18:58 positive positive trajectories and
18:59 they're getting really cool jobs so it
19:02 it's super important uh am I excited
19:05 about code interpreter [ __ ]
19:07 yes now I got it today so so I was
19:11 actually speaking in front of a oh this
19:13 was wild so today I did a an hour and a
19:17 half I did a 90minut lunch and learn for
19:19 this New York City Data visualization
19:22 company right it's a it's a good friend
19:24 I used to work with at agency.com and he
19:26 invited me to speak to the to the group
19:29 to his company and they do data
19:32 visualization five minutes before that
19:34 call I logged into chat GPT and I said I
19:37 wonder if I've got code interpreter and
19:38 there it was so so I hopped on this call
19:41 I did my sort of talk for for 90 minutes
19:44 and then the last five minutes I said
19:46 let let me let me show you something and
19:48 I showed them code
19:49 interpreter this is a company that does
19:51 this [ __ ] for a living and they were
19:53 like they were just like [ __ ] blank
19:56 blank
19:57 stairs like
20:00 because it was funny like when I started
20:02 the call I had everyone go around and
20:03 say what they were doing they were like
20:04 well yeah I think AI is okay but I'm
20:07 taking a little wait and see attitude
20:09 they all had attitude they were New York
20:10 you know New Yorkers I love New Yorkers
20:12 they had a little attitude by the end of
20:14 the call they were like uh uh I think we
20:17 got to deal with this so that was pretty
20:20 fun but yes I'm very excited about code
20:22 interpreter I just I saw I saw a Reddit
20:24 post um right before I came live that
20:27 that basically said any anyone else
20:29 feeling stupid today about code
20:31 interpreter cuz it's like
20:34 it's like I don't know [ __ ] about data
20:37 analysis so like I don't know how to get
20:39 data sources once I get a data source I
20:42 don't know what to do with it I don't
20:43 even know what questions to ask
20:45 it so I'm feeling I'm feeling kind of
20:50 like I felt when I was trying to get
20:52 stable diffusion running for the first
20:53 time where I just felt like a [ __ ]
20:55 [ __ ] for like
20:57 weeks and then I and and then he did the
20:59 same thing when chat GPT came out I'm
21:00 like I like I was doing parlor tricks
21:02 like having it write poems and songs and
21:04 I'm like it probably does more right and
21:08 I just felt stupid for like
21:10 months and then and so now I feel the
21:12 same way with code interpreter but yes
21:14 I'm very excited about it does chat GPT
21:16 read tarot cards it does and in
21:21 fact let's go hold I'm going to show you
21:24 a cool
21:27 thing
21:30 we're going to go um chat
21:35 GPT tarot card with
21:41 images a prompt that generates tarot
21:43 card
21:44 reading so if
21:46 you uh let's see what site is
21:50 this is this it yes chainbrain
21:55 ai.com
21:57 chain chain brain AI you can make money
22:01 with CHP come to chain brain AI you can
22:04 make money with ch okay so this is just
22:07 a couple of interesting prompts but one
22:09 of them is tarot card reading with
22:11 embedded images so I'm I'm going to show
22:13 you the
22:14 prompt well I'm not going to show you
22:16 the to it there we're going to go to
22:17 chat GPT we're going to play with
22:22 it oh do I need no I don't I don't think
22:26 I
22:27 do we'll do gp4 I don't think I need it
22:30 connected to the internet to do this or
22:32 maybe I do let's
22:36 see okay so what this what this prompt
22:39 does is it says you're going to act as
22:41 an expert tarot card reader generate a
22:43 tarot card reading with tarot cards from
22:45 major and minor minor Arcana follow the
22:48 following steps random number between 1
22:49 and 78 and then it's like insert random
22:52 number here insert the name of the tarot
22:54 card that responds to that number and um
22:58 do not use code block and then it it
23:00 points you to a a a Wiki media thing so
23:04 what this prompt has in it is um all of
23:09 the names of the files on Wiki media for
23:13 the
23:15 images and then you basically just hit
23:20 return and
23:25 it pulls your tarot cards for you and
23:28 and displays the images it's not
23:30 generating these it's literally what the
23:31 prompt is doing is it's using markdown
23:34 to point to a web image and then it's
23:36 just displaying the web image in your
23:38 browser so it's not
23:41 actually doing the stuff but it's a
23:43 tarot card reading right it picks three
23:45 cards and what's cool about this is you
23:49 know you can go back to that original
23:50 prompt if you know anything about tarot
23:52 card
23:54 reading you you could in instead of
23:56 doing like a three card pick you you
23:58 could do a nine card pick with two from
24:00 the bottom of the deck and you know all
24:01 the sort of fancy [ __ ] you see on
24:03 YouTube um so you could really bling the
24:06 [ __ ] out of this prompt and then yeah
24:08 and then it pulls it pulls these images
24:11 and if you don't like these images I bet
24:14 that there are other images somewhere
24:16 else on the internet like there's
24:17 there's is at Wikimedia right which are
24:20 kind of Open Source images but if you
24:22 found another place on the net that had
24:25 different kind of tarot cards you could
24:27 just swap out this
24:29 URL with with the the sort of names of
24:31 the individual cards down below isn't
24:33 that wild like that that's the sort this
24:36 is the sort of [ __ ] if like if you want
24:38 to if you want to
24:40 like really start to get your head
24:43 around prompt engineering and stuff like
24:45 this this is why you hang out on Reddit
24:47 in the in the chat GPT
24:49 forums because when when Geeks get their
24:54 [ __ ] heads around what's possible
24:57 with this there's like like the the the
24:59 depth of the rabbit holes that you can
25:01 run down with chat GPT are [ __ ]
25:10 insane all right I hope I'm
25:17 back why are you so horse Kyle I don't
25:20 know maybe it's because I talk on the
25:22 internet for 3 hours
25:25 straight why can't you walk straight my
25:27 ass is
25:34 numb you thought that was just a joke
25:36 Liz does chat jpt read tarot cards you
25:39 didn't think I'd have a [ __ ] answer
25:40 for that one did
25:42 you it
25:46 does I used to get on y'all's BBS which
25:50 BBS back in the olden timey days on
25:54 Echo all right we need more oh more
25:58 likes to boost the live guys support
26:00 thank you for that uh boost by tapping
26:02 the screen tap away people I encourage
26:05 the tapping I like the tapping the
26:07 tapping is good no one's giving me a
26:09 damn cowboy hat tonight what am I
26:11 supposed to
26:12 do do you think this is all caps people
26:15 do you think we will be able to download
26:17 an AI software and use on our own GPU
26:20 for
26:22 processing yes finesser we will be able
26:24 to
26:25 download on our own GPU okay so here's
26:28 what's
26:29 happening right now if you want to run
26:31 something the size of chat GPT you need
26:37 some [ __ ]
26:38 Hardware like George Hots the guy that
26:41 from Kama the uh the the full
26:44 self-driving car competitor to Elon Musk
26:47 he's a he's a weird but incredibly smart
26:51 dude he's started a company called Tiny
26:55 Box where he's selling he he he's
26:58 designing right now $115,000 computers
27:02 that have six gpus in them so you can
27:04 run this [ __ ] natively so if you got 15
27:07 grand yes you can just [ __ ] fire that
27:10 [ __ ] up um what's happening in the open
27:14 source Community is they're taking much
27:16 smaller models so so uh GPT 3.5 is 175
27:21 billion parameters I don't know what the
27:23 [ __ ] that really means it means it's big
27:25 it's a lot the llama models that were
27:29 released into the wild the the the core
27:31 models that meta released there was a 7
27:34 billion a 13 billion and then a 35
27:37 billion or 30 I don't know like whatever
27:39 65 billion whatever they they're they're
27:42 smaller what the op Source Community has
27:44 done is they have trained those smaller
27:47 models by getting chat GPT to create
27:50 training data and then and what's
27:52 happening is chat gpts 4 is creating
27:55 this training data like lots of it
27:58 they're training these small smaller
28:00 models and these smaller models are
28:01 getting very smart very fast for like a
28:03 100 bucks in
28:04 training what the developer Community is
28:07 now doing is they're taking those models
28:09 that have good performance and they're
28:10 optimizing them they're figuring out how
28:12 much data they can pull out and still
28:13 have these things function at a high
28:15 level so I saw so so there are now open-
28:19 Source models you can install on a phone
28:21 there are open source models that run
28:23 natively on an M1 Mac and an M2
28:26 apparently the M2 Macs are [ __ ]
28:28 destroying it um so so I think we are
28:33 probably easily within a year of being
28:36 able to do that
28:38 relatively um
28:41 seamlessly so yeah it's
28:43 coming U but if you want to run this big
28:45 [ __ ] the big models you got to have some
28:47 you got to have some iron as they
28:54 say chat GPT is 70% nerfed I don't
28:58 uh disagree with that I think we we are
29:01 not see well so like there's a bunch of
29:02 stuff you can't do um you can't do the
29:05 image generation it doesn't have the
29:07 32,000 token limit there's there there's
29:09 a bunch of stuff that I have a feeling
29:13 we're not
29:15 seeing a lot of what this model's
29:19 capable of I I think you're
29:23 right and I think they're slow rolling
29:25 it on purpose I think they know that
29:30 that they h i i I think they know that
29:33 they have to have it in the
29:34 wild long enough to discover [ __ ] like
29:37 they discovered with browse with Bing so
29:40 so if you don't know this three days ago
29:41 they pulled GPT connected chat GPT
29:44 connected to the internet they pulled
29:46 that functionality because what they
29:48 learned probably from a lawsuit or a
29:50 nasty call from a lawyer was that chat
29:52 GPT was pulling content from behind pay
29:55 walls without paying for them Publishers
29:58 apparently get a little cranky when that
30:00 happens um so yeah so I have a feeling
30:03 they're trying to figure all this [ __ ]
30:05 out before they really launch the the
30:07 powerful version of this and who the
30:09 [ __ ] knows what chat gp5 is that what
30:13 the core model of GPT 5 is it's going to
30:16 be
30:16 insane um you
30:19 switching your favorite for the first
30:23 company to make a personal agent
30:26 ubiquitous you switching your favorite
30:28 for the first com oh
30:32 oh am I like will I switch to anyone
30:35 that makes the first personal agent
30:37 ubiquitous I'll switch to anyone that
30:39 makes a personal agent functional
30:41 absolutely in fact this list of like the
30:45 only reason this
30:47 list I mean if you've been watching my
30:49 lives I've been doing these for what a
30:50 month something like that this list
30:52 hasn't changed I will change this list
30:55 the minute someone topples gp4 it'll go
30:58 to the top of the list um the minute
31:01 someone comes up with an autonomous
31:03 agent that doesn't suck that's that's
31:05 actually a personal agent like that that
31:07 Pi thing I'm I'm keeping my eye on them
31:09 I've got I've got my eye on anthropic
31:12 and you know anthropics here in
31:14 po um and and inflection I think it's
31:18 called Uh that's got this Pi personal
31:21 assistant um the pi personal assistant I
31:23 think it's called I'm keeping my eye on
31:25 them cuz that's the guy that founded
31:28 Deep Mind and deep mind's the one that
31:30 did Alpha fold that is [ __ ]
31:33 predicting um protein folding structures
31:36 with remarkable accuracy in the hundreds
31:39 of thousands
31:42 like Wicked smart folks so so yeah if if
31:46 you're asking will I change my alliances
31:48 in a heartbeat [ __ ] yes all this stuff
31:51 is Chang like everything that's
31:53 state-of-the-art today is going to be
31:54 laughably bad we're going to look back
31:56 at it and go remember when we had to
31:58 type prompts in the chat GPT that's
32:02 hilarious the kids will be like you used
32:04 to have to type what are you talking
32:07 about
32:10 Mama co-pilot is already in VSS code and
32:13 it's pretty good interesting very cool
32:16 laughing my ass off I caught myself I
32:19 caught up myself by commenting
32:22 good why Nerf the system
32:25 Xbox um AI versus educ oh versus Ed
32:31 Educators or our universities done
32:35 um I think universities that pretend
32:38 that this [ __ ] isn't transforming
32:40 education are done I think any
32:42 Universities that are not like oh we got
32:45 to change [ __ ]
32:47 everything I think are done are in
32:49 trouble yeah I
32:52 do any University that's still got a you
32:55 know you can't cheat you know we're
32:57 we're going to have anti-chat GPT
33:02 sniffers are done if they're if they're
33:05 not if they can't look at this and
33:08 recognize everything's fundamentally
33:10 different they're they're [ __ ] um what
33:13 was it last week Harvard announced that
33:16 their their sort of Flagship cs50 their
33:19 computer science their programming um
33:21 like their entry-level programming class
33:23 at Harvard is going to have a chat GPT
33:26 teaching assistant
33:28 so you'll have 24/7 access to an AI
33:32 teaching assistant for the computer
33:33 programming course that will check your
33:36 work help you with documentation debug
33:39 [ __ ] and then if the AI teaching
33:44 assistant can't help you I suppose you
33:45 can go to office hours with the real
33:47 person with the real Professor so that
33:50 like that's Harvard leaning in the the
33:53 New York City School District initially
33:56 banned chat GP
33:57 and reversed that the city of Boston put
34:00 out guidelines for gener generative AI
34:03 so you're starting to see there there
34:05 was a there was actually a newspaper
34:07 article in the New York Times probably
34:10 in January or
34:12 February that interviewed a bunch of
34:14 University provosts and you know
34:17 curriculum people um that were actually
34:19 talking about we've got to take this
34:21 seriously we've got to rethink education
34:23 so there's actually you know there's
34:25 movement in the in the higher space to
34:28 deal with this but there's also just
34:30 like staggeringly idiotic
34:33 close-mindedness to
34:35 it and you know that that's
34:38 just universities are big institutions
34:41 and big institutions hate change and
34:44 this [ __ ] is forcing change way faster
34:47 than I think any of us are prepared for
34:50 way
34:53 faster I'm amazed at VBA coding through
34:57 open Ai and other programming languages
34:59 wow I know it's amazing isn't it I know
35:02 I know I know this what I'm this is
35:05 like have your Kevin McAllister moment
35:08 right like use these tools until you
35:10 have that moment where you're like here
35:12 here here's the
35:16 thing chat
35:18 GPT the the box that you type in looks
35:22 like the Google box this ain't Google
35:26 it's not a Google search
35:29 in fact it's kind of the opposite of a
35:31 Google search where Google wants you to
35:34 be as concise as possible just give us
35:36 the minimum amount of information you
35:37 need in that box and then we'll figure
35:40 all the other [ __ ] out chat GPT is kind
35:43 of the opposite where if you give it too
35:44 little it just gives you crap back if
35:47 you start interacting with it not like a
35:50 search engine but more like a
35:52 collaborator like a like an intern like
35:55 a really [ __ ] smart intern that
35:57 that's clueless about
35:59 life that's chat GPT brilliant like
36:02 Savant level brilliant knows all the
36:05 [ __ ] and literally can't figure out how
36:09 to like take the next step so act like
36:12 someone who knows what they're doing
36:15 okay I want you
36:18 to okay here you go oh no that's
36:21 completely wrong okay sorry about that
36:25 want me to do it again
36:28 it's like a simpleton chat GPT is like a
36:30 simpleton that knows [ __ ]
36:35 everything really is and like the minute
36:38 you the minute you sort of figure that
36:39 [ __ ] out like something
36:43 shifts you're like holy [ __ ] I've got
36:46 like a savant idiot
36:52 partner I don't know it's I don't know
36:55 it's good it's fun it's good it's
36:58 awesome by the way there's 100 people in
37:01 here it's Friday night shouldn't you be
37:03 out like drinking or dancing or clubbing
37:05 or something not sitting with like oldie
37:08 mcart face in the in the in the in the
37:11 AI in the in the in the smartass MC AI
37:16 mcon all right I'm trying to find where
37:18 I was in the comments because Tick Tock
37:21 is it's a conspiracy theory where Tik
37:24 Tock doesn't want me to answer all of
37:26 your
37:27 comments but I refuse to let the man
37:32 keep me
37:34 down all right there's the VBA
37:38 comment your broadcasts are fantastic
37:41 thank you Firefly I appreciate that
37:43 alignments is what humans have to do
37:45 that's to a great degree that's right
37:47 right this is not it's one of my pet
37:49 peeves is like people are like the
37:51 robots are stealing my [ __ ] the robots
37:53 are doing this the robots wrote this the
37:55 robots wrote that the robots lied no
37:59 human beings asked the robots to do
38:01 something and then the human beings
38:03 didn't check their work and the human
38:05 beings are
38:07 idiots so so yes I agree you know humans
38:11 need to get their [ __ ] together and get
38:13 getting AI literate is probably the
38:16 single best thing any of us can do for
38:19 the next
38:20 year as as scary as it is and as
38:23 technical as you think you're not you
38:25 don't have to be technical with this
38:28 there's there's there's a bifurcation
38:29 that happened on November 30th before
38:33 November 30th if you wanted to play with
38:36 machine learning you needed the
38:38 propeller hat and the math degree right
38:41 and you needed to be a coder and an
38:43 engineer and you needed to live over in
38:44 that side chat GPT marks an inflection
38:49 point in the history of
38:52 humanity that we all anyone who can
38:55 basically get onto a machine if you can
38:57 get on to Tik Tock that's way more
39:00 [ __ ] complicated than getting onto
39:02 chat GPT so you can get on to Tik Tok
39:05 you can go here and we are like this is
39:08 the way I think about it I feel like I
39:11 feel so incredibly lucky to live at a
39:15 time in history where we get to play
39:17 with tools this powerful like we meaning
39:20 everyday people not machine engineers
39:23 and
39:24 scientists all of
39:25 us so wherever you are on the spectrum
39:28 of geeky to
39:31 clueless you can do this prompt. chat
39:35 will teach you how to do this chat.
39:37 open.com is free bing.com is free po.com
39:41 is
39:42 free go play with those things until it
39:45 blows your mind and then get then dig in
39:49 only people have access to the
39:52 unregulated version can speak about this
39:55 I don't know what that was in references
39:56 to
39:58 um humans will have to change humans are
40:00 going to have to adapt absolutely Skynet
40:03 okay somebody saw a movie Sky not the
40:07 robots are going to kill us the robots
40:08 are going to kill us I'm sorry Dave I'm
40:10 afraid I can't do
40:15 that they might kill
40:18 us what
40:20 if they make make life dramatically
40:23 better that is a plausible future as
40:26 well it's just that's a more boring
40:28 movie so that's why we see 500 movies
40:32 where the robots kill us and like six
40:34 where the robots are
40:36 sweet nobody wants to see a movie where
40:39 the we we hug and the robots Usher us
40:43 into a a a a great Renaissance of of
40:48 creative expression the likes of which
40:50 the world has never seen no one would
40:52 want to see that
40:53 movie it's a possibility though if we
40:57 choose to make it so in the mental
41:00 health field I'd give it a Master Level
41:03 PhD understanding
41:06 interesting I think there's something
41:09 there's something really like like just
41:11 kind of ironic and weird about the fact
41:13 that the large language model literally
41:15 doesn't know what it's saying it's it it
41:18 is it's got the it's it's got
41:21 the each word that it types it's doing
41:24 it sort of with the functional
41:25 equivalency of a c calcor right so if
41:28 you type in 4 plus 4 in a calculator and
41:30 it spits out four that's what a large
41:33 language model is doing word by word
41:35 it's just spitting out a probability
41:37 this word then this word then this word
41:39 then this word it's not aware of
41:41 anything it's writing
41:43 but it's so good because it's been
41:46 trained on what it's been trained on and
41:47 they've got it fine-tuned how they've
41:49 got it
41:50 fine-tuned it's it behaves remarkably
41:53 well with empathy and with nuance
41:57 and with engaging so so yeah I think
42:01 companionship and um therapy a and just
42:07 mental health in general and and
42:09 caregiving in general I think that these
42:12 large language models are going to be
42:14 [ __ ] brilliant I think they're going
42:15 to make doctors better I think they're
42:17 going to make caregivers better I think
42:19 they're going to augment the health care
42:23 industry and yet they're completely
42:25 clueless but there was a study I don't
42:27 know a month or so ago that a bunch of
42:29 patients interacted with chat GPT
42:32 doctors you know via text and real
42:35 doctors via text guess which one they
42:37 liked
42:38 more it wasn't the real
42:42 doctors more competent more empathetic
42:46 that was the feedback the chat GPT
42:48 doctors more competent more empathetic
42:51 it's like holy
42:54 [ __ ] maybe these
42:57 tools will do all the shitty ass
43:00 Insurance paperwork that makes doctors
43:02 so cranky and makes them not spend time
43:05 with patients so maybe it will free them
43:07 up to spend more time with patients and
43:09 maybe it will actually train them to be
43:12 better doctors like better bedside
43:15 matter better empathy I think that's
43:19 coming I got my floating screen back I
43:21 know I did they gave it back to me I was
43:24 a good boy oh [ __ ] I have to see if if I
43:26 can enable code interpreter yeah go go
43:29 go if you've got chat GPT plus go to
43:32 your settings and go to Beta beta apps
43:35 or beta beta features and see if code
43:37 interpreter is there and if it's there
43:39 click on it turn it on go play I'll I'll
43:43 show you how to do that here in a second
43:44 if you want um expand what did you do
43:49 expand what did you do what did you do
43:52 what did you do bad dog bad dog I don't
43:57 know what I did but I was very very very
44:00 bad it's amazing when you talk to people
44:03 about AI there's a real division between
44:05 people on AI there is that's why this
44:07 channel
44:08 exists that's why this channel exists
44:10 thanks Ron
44:15 um everyone that's pissed off about it
44:18 that I know and everyone that dismisses
44:19 it and everyone that demonizes it hasn't
44:22 [ __ ] used
44:25 it
44:27 it's why I start out and mention 30
44:31 times a [ __ ] live go here and play
44:35 with these tools if you haven't played
44:36 with chat gbt go here and play with that
44:40 until you understand that this is not a
44:42 Google Search and this is
44:46 not a toy and it's not
44:50 hype it is something
44:54 different it is a reflection of humanity
44:58 that we get to tap
45:00 into and it's a Humanity amplifier it
45:03 can take this [ __ ] can
45:06 take your little Whisper of an idea that
45:10 you're like you know maybe someday I'll
45:13 fill in the blank write a novel write a
45:15 play write a recipe start a
45:19 garden start a community group start a
45:22 church whatever the [ __ ] you want to do
45:23 that little whisper that you've always
45:25 kind of avoided cuz you're like I don't
45:27 really know how to do it I don't know
45:28 where to
45:30 start go
45:32 there and just start [ __ ] with it if
45:36 you haven't done
45:37 it have it write a poem have it write a
45:41 song have it write a project plan have
45:43 it write a marketing brief have it write
45:45 a
45:45 blog and then and then at that point
45:48 you'll be like
45:50 huh that sucks less than I thought it
45:53 would but surely it can't do what I do
45:58 and then it will some version of what
46:00 you do and then you should be freaked
46:03 out appropriately you should be
46:05 absolutely [ __ ] terrified
46:08 and you should be inspired to go holy
46:11 [ __ ] I got to deal with this I got to
46:14 figure this [ __ ] out then We're Off to
46:17 the
46:19 Races so yes I agree so and and you know
46:23 what I would say is for those people
46:25 that are violently opposed to this and
46:28 angry about it like have empathy for
46:32 them because it's [ __ ] scary and and
46:35 if you
46:37 know especially for people that you know
46:41 are on in years and maybe have 20 30 40
46:46 years into a
46:47 profession and all of a sudden [ __ ]
46:51 like me are going you might want to
46:52 reconsider your career choice well I'm
46:54 [ __ ] 30 years into it
46:58 jerk so you
47:02 know you know have empathy and then I
47:04 think it's I you know I think the thing
47:05 to do is is do everything you
47:09 can to make what you know about this
47:13 stuff accessible to them in a way that
47:16 it you know they can sort of see
47:19 themselves in in in it right do
47:21 something for them say Hey you know
47:24 weren't you trying to to come up with
47:27 some book ideas well yeah but I didn't
47:30 really get it started hey you want to
47:32 you want to try something what what was
47:34 the what was the idea about type that
47:36 into chat GPT did do you have an outline
47:39 well I don't really know how to do an
47:40 outline let's let's see if we can get
47:41 this thing to make an
47:43 outline you know give me three different
47:45 story Frameworks that I could you know
47:48 write my novel around okay here they
47:51 come oh wow and and and what's the
47:55 character's name name Bob okay let's put
47:57 Bob in there and what's what's Bob do is
47:59 it mechanic okay we'll put that in
48:02 there write me the outline of Bob who
48:05 finds his true love you know at the ball
48:08 game and whatever the plot points are
48:11 out comes the outline then give that to
48:15 them that's pretty cool that's that's
48:18 not as scary as I thought you know try
48:21 to get people
48:23 there open AI just posted positions for
48:25 their alignment Team all north of
48:29 $400,000 there you go wow Jesse really
48:32 holy [ __ ]
48:35 $400,000 starting starting at
48:39 $400,000 now if you're going to go work
48:41 at open AI for their alignment team I
48:43 assume you got to have some [ __ ] Tech
48:44 chops I assume you got to have your [ __ ]
48:46 going on with this here uh this here uh
48:50 code
48:53 interpreting but I don't know it could
48:55 just be Charming good haircuts that get
48:57 you in
49:00 there I guess I'm out
49:03 [ __ ] I had this idea which one laughing
49:07 my ass off at the mental image of Kyle
49:09 deep in the YouTube rabbit hole of tarot
49:11 card dealing
49:13 exactly can you share this can I share
49:15 what I'm sharing everything I don't know
49:18 what uh Joe famous good name by the way
49:20 Joe famous that's solid I like it and
49:22 then right below Joe famous is Johnny
49:25 bullets this this is awesome tonight
49:28 except for the troll that told me I was
49:29 going to be permanently banned Jesus
49:31 can't believe I fell for that what next
49:33 a Nigerian
49:35 prince oh that's the other thing the
49:37 Nigerian prince letters are going to get
49:39 much better written you know that right
49:41 okay um so I don't know what you were
49:43 talking about Joe famous I I can share
49:45 whatever if I know what it is um where
49:49 do you go to make it to video for free
49:53 to make what to video to to like do text
49:55 to video video um you can go to Runway
50:00 ml um are those free I think they are
50:05 free I don't remember anyway go to
50:08 Runway ml they have an app and then they
50:11 have a Discord server where you can
50:13 generate you can either do video to
50:15 video or text to
50:17 video I I think that's what you're
50:19 asking Johnny bullets but I'm not
50:21 sure what are the cool what are the cool
50:25 jobs out there for AI newbies
50:32 um I think every single company within
50:36 within probably before the end of the
50:38 year but you know let's let's call it
50:40 within a year let's get let's give
50:42 people another year we're in July let's
50:44 say by this time next
50:46 year every single
50:50 company is going to need a significant
50:53 amount of AI literacy in in in their
50:59 organization um and they may already
51:01 have it but not know it um go go check
51:03 out there's an article from a guy named
51:05 Ethan mullock called detecting the
51:07 secret cyborgs where he's talking about
51:09 there's a class of AI literate people in
51:12 companies right now that are using AI
51:15 for personal productivity like dramatic
51:17 improvements in personal productivity
51:20 but they're not telling their companies
51:21 about it they're the secret cyborgs and
51:23 the companies are like well these tools
51:25 aren't ready for prime time and we don't
51:27 have a privacy policy and we're not
51:29 really sure what we're going to do with
51:30 AI and so the organizations are are kind
51:33 of like trying to figure out if and when
51:35 they're going to start doing this
51:37 meanwhile they've got a whole class of
51:38 employees that are AI literate and
51:41 getting increasingly so um and and he's
51:44 saying basically bridge the gap
51:47 so so what I would say is rather than
51:51 what are the cool AI jobs out there you
51:53 know there's there's going to be prompt
51:55 engineer jobs and there's going to be
51:56 just all sorts of thing I would say go
51:58 back to yourself what are you good at
52:00 what are you passionate about get AI
52:03 literate and start talking about it
52:05 start talking about it on LinkedIn start
52:07 talking about it on Facebook start
52:09 talking about it at [ __ ] dinner
52:11 parties start annoying people like
52:14 people can't stand talking to me anymore
52:16 I'm convinced of that cuz all I [ __ ]
52:18 talk about is this stuff why cuz it's
52:20 going to change the world and I want to
52:22 be a part of it and I want to bring
52:24 people along it's a [ __ ] adventure to
52:25 me me um
52:29 in getting AI literate and
52:32 in communicating it with the world you
52:35 will become
52:36 known within your
52:39 circles as one of those people that when
52:42 they finally when the light bulb finally
52:44 goes off for them and they're like I got
52:45 to learn about this AI oh you know what
52:50 um what's your name Saku Sal Saku
52:56 um Sally knows about
52:59 that Sally's been talking about that for
53:01 like two months it's all it's all they
53:03 talk about hey Sally Could you teach me
53:07 this like be that person so
53:11 so so I don't know I mean start paying
53:13 attention to job boards um within the AI
53:17 Salon the the salon group that I
53:20 created
53:21 um a number of people within there came
53:25 into the group not knowing anything
53:26 about AI other than having a
53:29 um curiosity about it started learning
53:33 about it started posting about it put it
53:36 on their LinkedIn profile started
53:38 posting articles and posting results
53:40 here's something I did in mid Journey
53:42 here's something I did in chat
53:44 GPT are getting jobs are getting cool
53:47 jobs are getting AI Centric jobs so I
53:50 would say just start learning start
53:52 putting yourself out there I'm already
53:54 running models on my RT TX 490 that's
53:56 super cool thanks Ron Burgundy that's
54:00 okay solid solid troll compliment sort
54:04 of Burn yeah I like
54:08 it don't Google TIN Tiny
54:12 package Tiny Box George hot's company is
54:16 called Tiny Box not tiny package you
54:19 should definitely do not Google tiny
54:21 package don't do I know you're doing it
54:23 right now you're all doing it right now
54:25 you're Googling tiny package don't do it
54:28 Jesse tried to warn
54:31 you GPT for video free no BS I go
54:39 try Runway ml but also just go to Future
54:54 pedia.com imagine any are are free free
54:57 maybe maybe you can get some amount of
54:59 credits each day what are the names of
55:02 those open source models oh the the the
55:06 core models that meta released are
55:08 called
55:09 llama and then one of the first ones
55:12 that was sort of trained up was called
55:14 alpaca and now there's one called vicuna
55:16 and then there's one called MPT 7B and
55:21 then there was MPT
55:23 7B which stands for 7 billion an MPT
55:27 7B Storyteller or story something that
55:30 had a 65,000
55:32 token context window and
55:36 like what I would do is I would I would
55:39 just go to Twitter and to
55:42 Reddit and type in op Source
55:46 llm both on Twitter and Reddit and just
55:49 run down those rabbit holes I can't keep
55:52 up with the open- source models anymore
55:54 the the base mod model that meta
55:56 released are called llama like the
55:59 animal um and then all these open source
56:03 projects are stacking on top of it and
56:05 there's just there's hundreds of them
56:06 now um and all sorts of different
56:09 flavors of them it's just it's the open
56:11 source community and it's it's the open
56:13 source Community where everyone has
56:15 basically you know dived in dove in dove
56:19 in head first cuz this AI stuff's so
56:22 exciting um so just go to Twitter open
56:25 source llm and redddit open source llm
56:28 and you'll you'll find you'll find
56:30 projects there that are people are
56:32 excited about they probably already have
56:34 five if you COI if I can explain to me
56:38 what chat GPT actually is I'd really
56:41 appreciate it um all right how about I
56:44 teach you how it works you want to know
56:46 how chat GPT works well so let me
56:48 explain what it is
56:51 okay now let
56:53 me h
56:57 I'll show you how it works in a second
57:06 [Laughter]
57:07 [Music]
57:10 but so you know how Google
57:16 is they indexed all of the internet they
57:19 went out and they made hyperlinks to
57:22 websites and they said this hyperlink
57:25 points to a website that talks about
57:27 this [ __ ] and then when you go say I
57:29 want a website about that [ __ ] Google
57:33 has somehow figured out how to search
57:35 all of the sites on the internet and
57:36 give you links that point you to those
57:41 sites chat GPT has learned not just the
57:44 link and what the site's about but it
57:47 basically goes behind it and it's
57:49 learned all the
57:51 [ __ ] at the
57:53 site so when you interact with it on its
57:57 surface it feels like kind of a Google
57:59 search box it's just a simple here hang
58:02 on let me show
58:07 you it's just a simple blank window with
58:10 a with a simple it says send a message
58:14 right but the message I can send instead
58:17 of getting back a list of links to other
58:19 sites I can put in here
58:23 explain the main concepts of
58:29 quantum mechanics as well let me just do
58:34 that the main concepts of quantum
58:37 mechanics and so you see how fast it's
58:40 writing so it it's writing now it's
58:42 explaining to me what quantum mechanics
58:43 are it's not taking me to a site that's
58:46 been SEO optimized it it it knows all
58:49 the [ __ ] it knows all the languages it
58:51 knows all the concepts it knows all the
58:53 [ __ ] so it's explaining to me
58:55 wave particle duality and superposition
58:58 and Quant quantization and uncertainty
59:01 principle and wave function and
59:02 probability and entanglement you know
59:04 all the cool stuff with quantum
59:06 mechanics but now I can say
59:09 um explain
59:12 it
59:13 like I am
59:16 five sure imagine you have a toy car
59:20 that can also act like a ball sometimes
59:23 when you play with it the car moves like
59:24 a car and the other times it rolls and
59:26 bounces like a ball so so in 10 seconds
59:30 that was it it shifted from explaining
59:32 quantum mechanics to explaining it to a
59:35 5-year-old now I can say um make it an m
59:40 m
59:43 WP first one yo listen up
59:47 kid I'll break it down real quick
59:49 quantum mechanics it's a mindbending
59:51 trick picture a toy car it's got dual
59:53 identities acting like a car bouncing
59:55 ball with intensities so it's
59:58 rhyming and talking sort of like Eminem
1:00:01 would talk to a 5-year-old would wrap to
1:00:04 a 5-year-old so now I can say um make a
1:00:08 table that
1:00:11 compares
1:00:14 the
1:00:16 Einstein um
1:00:19 relativity world and the quantum
1:00:24 world
1:00:26 and now it
1:00:27 will
1:00:29 hang there you
1:00:31 go it makes a table comparing two fairly
1:00:35 deep
1:00:36 subjects Einstein relativity relativity
1:00:40 macroscopic Quantum World microscopic
1:00:42 subatomic
1:00:44 right
1:00:46 um write
1:00:49 a python
1:00:54 application
1:00:56 application that um
1:01:01 takes some inputs and
1:01:05 simulates
1:01:07 both physics
1:01:10 models sure here's here's a python
1:01:14 script here's a python application that
1:01:16 allows you to input values and Sim
1:01:18 simulate both Einstein relativity world
1:01:21 and the quantum world the example
1:01:23 focuses on basic calcul ation and
1:01:25 simulations rather than comprehensive
1:01:27 models you can modify and expand upon it
1:01:29 and then there's your code
1:01:32 so what I just did is what I consider
1:01:35 the parlor tricks of chat
1:01:38 GPT but if you think about what I just
1:01:41 did and what it just did it is
1:01:44 staggeringly remarkable right this is
1:01:47 not a Google search at
1:01:50 all this is us being able to tap into
1:01:55 to effectively the collective knowledge
1:01:58 that's been put on the internet for the
1:02:00 past five decades the public internet
1:02:03 anyway and do it instantly and do it
1:02:06 exactly as we want it and and get like
1:02:09 be able to bend this tool to our will so
1:02:12 so what it is is a tool that allows you
1:02:15 to kind of do anything you can imagine
1:02:19 when it comes to knowledge you could
1:02:20 learn something you don't have to learn
1:02:22 programming it will program for you it
1:02:25 will teach you how to use its
1:02:27 programming that it creates to actually
1:02:30 do something you want you know if you
1:02:32 want to you want to be an app developer
1:02:35 but you've never learned how to program
1:02:36 you can actually do
1:02:38 that going to take you some
1:02:40 work but it's kind of here to teach you
1:02:43 to do that
1:02:44 so the let me let me flip the
1:02:53 screens the thing that makes it more
1:02:56 remarkable is how it does what it does I
1:02:59 don't had it partner you know what we're
1:03:00 talking about here oh
1:03:06 hey you know I kind of like that their
1:03:08 internet you ever been on that internet
1:03:10 they got this new thing it's called
1:03:13 AI
1:03:17 um
1:03:21 cheers the way it actually works
1:03:25 is when you type in your
1:03:28 prompt some [ __ ] Geniuses out there
1:03:31 Jeffrey Hinton and the people that made
1:03:33 the Transformer there's a bunch there's
1:03:34 a bunch of geniuses out there what they
1:03:36 figured
1:03:38 out they they figured out something that
1:03:40 you've already played with so the AI
1:03:42 already exists in in almost all the
1:03:44 software we have you know when you're in
1:03:45 Google sometimes you're writing an email
1:03:48 and you write a sentence and Google will
1:03:51 put those gray words at the end or
1:03:53 you're in Microsoft Word or where it
1:03:55 completes your
1:03:56 sentence that's essentially what chat
1:03:58 GPT is but it's like that on
1:04:01 steroids so when you write your prompt
1:04:04 and you submit it what it does is it
1:04:06 turns that into a
1:04:08 mathematical probability
1:04:11 engine that says okay based on all of
1:04:13 what this person just wrote what is the
1:04:16 next most probable word that is the ne
1:04:19 that's the right word and it and it puts
1:04:21 one word there and then it takes all of
1:04:24 that plus one word and it says well
1:04:25 what's the next most probable word and
1:04:27 what's the next most probable word and
1:04:29 it does that over and over and over so
1:04:31 when you saw it just spit out here's
1:04:33 here's quantum mechanics explained
1:04:35 here's the poem here's the code it's
1:04:38 doing that one token at a time they call
1:04:40 it a token it's it's part of a word it
1:04:42 doesn't matter but just think one word
1:04:43 at a time over and over and over again
1:04:46 just a probability engine so so
1:04:49 conceptually it's quite simple what it's
1:04:51 doing but realistically because it's
1:04:53 been trained on [ __ ] everything and
1:04:55 because it's so
1:04:57 fast it's it's like this magic access to
1:05:03 the knowledge of the of humanity you
1:05:05 know some large percentage of it so I
1:05:08 hope that explained
1:05:09 it the raw gp4 model was aoral but fast
1:05:14 and vastly smarter than it is today yep
1:05:17 and it was well it was it was
1:05:19 unremarkable it got remarkable as they
1:05:21 fine-tuned it and then they had to put
1:05:23 guard rails on it to to you
1:05:25 know keep it from doing some bad [ __ ] so
1:05:29 like I you know I don't listen as as
1:05:32 hobbled as gp4 might be if you [ __ ]
1:05:35 use it on a regular basis it's still
1:05:37 magic it's still a remarkably powerful
1:05:40 tool like in the hands of just Everyday
1:05:44 People this is the most powerful tool
1:05:47 ever put in the hands of normal people
1:05:49 it's you know I I I say it on here a lot
1:05:52 but I think it's Carl Sean was it Carl
1:05:54 Sean
1:05:55 I think it's Carl San wait
1:06:04 hold who
1:06:13 said Arthur C
1:06:15 Clark any sufficiently advanced
1:06:18 technology is indistinguishable from
1:06:21 Magic that's what this
1:06:23 is it's an advanced technology it
1:06:26 couldn't happen without these
1:06:27 supercomputers and without the crazy ass
1:06:29 math that some people figured
1:06:32 out but what it does is
1:06:35 Magic you ask it to write a rap or a
1:06:38 country song or your term paper or your
1:06:41 email or your resignation letter from
1:06:44 your shitty ass company that you want to
1:06:46 leave but you don't want to burn a
1:06:48 bridge and it [ __ ] does it like
1:06:51 that and it does it good it doesn't
1:06:54 probably better than you would it gets
1:06:56 you to 80% in 30 seconds and then you
1:06:58 can take it for the for the rest of the
1:07:01 20% do you think we would be able to
1:07:03 share GPU power to run GPT like people M
1:07:09 Bitcoin well that's that's exactly what
1:07:13 Bing is and that's exactly what chat GPT
1:07:15 is so all the free tools here chat GPT
1:07:19 and Bing are
1:07:21 that because every queer that you
1:07:26 do it's like it's burning you know
1:07:30 multiple gpus to be able to answer your
1:07:33 dumb little question right my dumb
1:07:35 little question like we we're doing that
1:07:39 right now
1:07:41 now are you saying that W will there be
1:07:44 Services where I
1:07:47 can take a powerful open- Source large
1:07:50 language model and and have shared GPU
1:07:54 Bank so I don't have to send my data to
1:07:56 the man and to Microsoft and Sam mman
1:07:59 yeah probably sure that'll
1:08:02 come you know unless the powers that be
1:08:05 are like yeah we're going to we're we're
1:08:07 going to cuz I think there was some
1:08:09 legislation somewhere where they're
1:08:10 trying to lock down the sale of gpus
1:08:12 we're not going to sell gpus to the
1:08:15 undesirables which is basically people
1:08:17 we can't make money off of so we'll
1:08:20 we'll see what happens with all the
1:08:22 legislation and regulation stuff um how
1:08:25 do they know that chat GPT doesn't pull
1:08:27 content wait how do they know that chat
1:08:29 GPT doesn't pull content from behind pay
1:08:31 walls they know that it does that's why
1:08:33 they puled that's why they removed it so
1:08:36 so pay walls work there there's kind of
1:08:39 two flavors of pay walls so when you go
1:08:42 to like the New York Times or you go to
1:08:44 some site and it says sorry you've got
1:08:46 to subscribe to
1:08:47 this one way you do that is kind of the
1:08:52 cheap way and if someone is Technic Tech
1:08:54 and they know how to inspect the code of
1:08:56 a website they can just inspect the code
1:08:58 and there's all the content so the pay
1:09:00 wall is just this sort of [ __ ] thing
1:09:03 there's another way you do pay walls
1:09:05 where it's actually behind a pay wall
1:09:06 and you can't just inspect the code and
1:09:08 see
1:09:09 it I
1:09:11 think browse with Bing if you had a
1:09:14 shitty pay wall that just you could
1:09:16 inspect the code and get the content I
1:09:19 think browse with Bing was just pulling
1:09:20 that content and saying here you
1:09:23 go but if you do if it's a big company
1:09:26 that had a shitty pay wall that big
1:09:28 company can still get pissed off and say
1:09:30 we're going to send all our lawyers
1:09:31 after you and destroy your company so
1:09:33 that's that's my guess we will laugh
1:09:36 that computers had keyboards yeah I
1:09:38 think so I mean I think if if you look
1:09:40 at the technology in in the Vision Pro
1:09:42 from Apple what's coming and
1:09:45 and yeah although you
1:09:50 know I don't know that typing goes away
1:09:53 because there's a lot of times when I'm
1:09:55 in a situation and I don't want
1:09:58 to um I don't want to
1:10:01 talk and I mean you know 10 10 years 8 8
1:10:07 10 years you can probably just think
1:10:09 thoughts and that's your interface so
1:10:11 you don't have to talk you can just
1:10:12 think it so that's probably coming don't
1:10:14 get me
1:10:16 wrong is the background on your laptop
1:10:19 the Sam
1:10:21 Alman I orb it's actually not a bad
1:10:24 that's not a bad thought this is an
1:10:26 image um
1:10:31 from from stable
1:10:33 diffusion that I created I have an art
1:10:36 project called Kyle Shannon
1:10:38 dreams and this was one of the images
1:10:42 that when I created it I I had a visc
1:10:45 visceral
1:10:48 Rea so it's like there's my you can you
1:10:51 can see my chin my little double chin my
1:10:54 lips my nose and then that thing on the
1:10:56 eye
1:10:58 um this was one that that when it was
1:11:00 created like when it showed up I I like
1:11:03 had this emotional response this is the
1:11:06 kind of art I would create if I had the
1:11:08 skills to create it from scratch and I
1:11:11 don't but I don't and so when I saw this
1:11:13 I'm like that's that's that's [ __ ]
1:11:16 magic like this is what I would create
1:11:19 um so that's why I use it but yeah if
1:11:20 you don't know what he's referring to
1:11:22 hang on let me flip back here
1:11:28 so Sam Alman has this company called
1:11:30 worldcoin and they have this
1:11:33 orb that you stick over your eye and it
1:11:36 scans your iris and it verifies that
1:11:39 you're
1:11:40 human and if you're willing to do that
1:11:43 Sam is going to give you some
1:11:45 cryptocurrency called
1:11:47 worldcoin and it sounds creepier than
1:11:50 [ __ ] but a lot of this stuff does and
1:11:52 what's going to happen
1:11:54 what what the purpose of it is is
1:11:57 imagine a Twitter or a threads you know
1:11:59 in today's
1:12:01 parlons where every person on there is
1:12:04 guaranteed to be a human and you know
1:12:07 they're human and and I can say oh
1:12:09 that's a video from the
1:12:11 verified you know Kyle and the
1:12:15 verification isn't $8 I pay to Elon Musk
1:12:18 but it's actually that's the verified
1:12:20 Kyle right and then you could imagine
1:12:24 built on top of that could be here's a
1:12:27 video that Kyle produced and and it's
1:12:29 blockchain verified that it hasn't been
1:12:31 modified and here's an article that he
1:12:33 wrote and here's a this and here's a
1:12:35 that and here's all the people that
1:12:38 commented on it and those were the
1:12:39 actual comments and nothing's been
1:12:41 manipulated so so the the orb thing is
1:12:44 is a very real thing trying to solve
1:12:46 what's going to be a very very real
1:12:48 problem that I think we're going to see
1:12:50 in the 2024 election here in the states
1:12:53 where uh
1:12:54 some shit's going down where if you
1:12:58 thought
1:12:59 misinformation and wacky High Jinks were
1:13:02 big in 2016 and 2020 you ain't [ __ ]
1:13:05 seen nothing 2024 is going to be insane
1:13:09 with Miss
1:13:10 information um the era of human
1:13:12 intelligence is already over I think
1:13:14 we'll get it back is Google search dead
1:13:16 Google search is [ __ ] struggling I
1:13:19 think they're gonna I think I think
1:13:21 they're going to have to reinvent
1:13:25 what it is I do I don't think it's dead
1:13:28 and I think it's I I think you know
1:13:30 there's there's there's an there's
1:13:32 enough of an
1:13:36 established infrastructure of search and
1:13:39 SEO and websites that they're going to
1:13:42 stay for a while but I think what
1:13:44 happens is the importance of destination
1:13:48 website starts to plummet like right now
1:13:51 Google's whole model is you search for
1:13:53 [ __ ] they give you you a link they want
1:13:54 you to click on 30 of those links and go
1:13:57 to a website what chat GPT is doing is
1:14:01 you ask for [ __ ] and it's there you
1:14:03 don't have to go
1:14:05 anywhere that's fundamentally a
1:14:08 different thing it's just a completely
1:14:10 different Paradigm and so I think it
1:14:12 makes the importance of a website less
1:14:16 important for end users cuz end users
1:14:18 could just say hey my knees hurt oh um
1:14:23 one of the things that might help is
1:14:24 minimalist running shoes oh what are
1:14:26 minimalist running shoes here's what
1:14:28 they are why do they make your knees
1:14:30 hurt less here's the science oh that's
1:14:33 kind of cool who are the companies that
1:14:35 do that here's the list of companies
1:14:37 that do that oh can I what are the shoes
1:14:39 look like here's what they look like oh
1:14:41 can I buy that one yeah here it is
1:14:42 here's the link buy it click it's on its
1:14:45 way I never left chat GPT or whatever
1:14:48 the tool is at the time it might be my
1:14:49 agent that goes out and does that for me
1:14:52 there's no need for me to go to a
1:14:54 website so if you play around with Bing
1:14:58 what you'll see is they're starting to
1:15:00 incorporate you know citations and links
1:15:03 and things into the answer
1:15:07 so it's probably still going to have
1:15:10 like SEO but it's going to be more like
1:15:12 llmo large language model thank you
1:15:15 thank you thank you large language model
1:15:18 optimization how do you get your company
1:15:20 to show up in that conversation about
1:15:23 minimalist
1:15:25 running I don't know someone's going to
1:15:28 have to figure that [ __ ] out Google
1:15:31 wants you to still go through their
1:15:32 portal to do all this but right now Bart
1:15:36 isn't as good as chat GPT and Microsoft
1:15:39 is subsidizing chat GPT and Microsoft
1:15:42 doesn't give a [ __ ] if they make money
1:15:43 on Advertising right now so it's in
1:15:49 trouble you know what did they call they
1:15:51 call it a red alert or something like
1:15:53 that
1:15:54 like four weeks after chat gbt came out
1:15:56 Google was like
1:15:58 H I know we invented this [ __ ] but uh
1:16:01 we're behind we're gonna have to do
1:16:03 something here anyway listen it's 11:45
1:16:06 here in Denver so if anyone's on the
1:16:08 east coast God damn you need to go to
1:16:10 bed how do I feel about uh solidarity AI
1:16:13 I don't know it I'll I'll look it up
1:16:15 though for the guy who wanted to make
1:16:18 the video with chat GPT check out the
1:16:20 real Melissa zando on Tik Tok okay cool
1:16:25 awesome I love that you're helping one
1:16:28 another hard to hear with the music not
1:16:32 sure what music we we got a bit of Jack
1:16:35 Nicholson Vive here's
1:16:40 Johnny yeah my hair's my hair's yeah
1:16:43 that's that's yeah yeah we got some jack
1:16:46 in there today yeah yeah yeah now we got
1:16:48 some jib carry all right Tom Arnold Vibe
1:16:52 combo nice
1:16:54 Jack see these are the kind of trolls
1:16:57 that I like what's the difference
1:16:59 between Ai and a quantum computer um AI
1:17:02 needs computers to run a quantum
1:17:04 computer is a quantum computer that can
1:17:07 process ridiculous amounts of data
1:17:10 um quantum computers aren't quite ready
1:17:13 for prime time yet but when they are uh
1:17:16 you're going to be able to process a lot
1:17:17 more [ __ ] a lot more quickly so I would
1:17:19 imagine that the ability to train these
1:17:22 models goes up dramatically and they get
1:17:26 powerful to the point that they probably
1:17:28 need to be licensed like you license a
1:17:30 nuclear power plant because they'll be
1:17:32 that powerful that's what Sam alman's
1:17:34 pushing for he's Sam alman's pushing for
1:17:36 three tiers of
1:17:39 Regulation the middle tier is kind of
1:17:42 where we are today gp4 Bard llama you
1:17:47 know the these foundational large
1:17:48 language models that are very very
1:17:50 powerful but they're kind of stupid
1:17:54 he's saying those the companies that can
1:17:56 produce the the base large language
1:17:58 models they should be regulated he's
1:18:01 saying any of those companies that reach
1:18:03 AGI or you know higher level cap
1:18:06 capability that should be licensed like
1:18:09 a nuclear power plant and then what he's
1:18:11 saying below that none of that should be
1:18:13 regulated all the small companies using
1:18:15 these tools should be able to [ __ ]
1:18:17 just go nuts and do what they do um so
1:18:21 it's going to be interesting to see how
1:18:22 it plays out
1:18:24 anyway everybody uh this was a good long
1:18:27 night I'm a little horse my butt's numb
1:18:29 you know it's a good sign it's time to
1:18:31 go can't feel my
1:18:33 legs but uh it's been a pleasure happy
1:18:36 Friday night or for happy Saturday
1:18:37 morning for some of you if you're on the
1:18:39 East Coast peace out everybody I will
1:18:41 probably see you sometime tomorrow I'll
1:18:43 be around you know I'll go fix the hair
1:18:46 and come
1:18:46 back
1:18:49 peace